Caixin Global published a cover story on 31 August 2026 arguing that the United States technology blockade has catalysed a boom in China's domestic AI chip sector. The report, by Gu Zhaowei, describes a sellers' market in which securing production capacity effectively guarantees demand. The full article sits behind a paywall, so this analysis draws on the portions Caixin made publicly available, including a detailed timeline of the sector.
The supply and demand gap
In the first quarter of 2026, domestic demand for AI computing power rose 417 per cent year on year, against supply growth of 128 per cent, according to figures from the state-backed China Academy of Information and Communications Technology cited by Caixin. Between April and July 2026, leading model developers including Z.ai, Moonshot AI and Alibaba paused or limited new subscriptions for individual users, a step the report attributes to the acute chip shortage.
The constraints follow tighter US export restrictions introduced in April 2025, which narrowed Chinese firms' access to advanced Nvidia chips. In the second quarter of 2026, deliveries of Nvidia's H200 chips to Chinese clients accounted for less than 1 per cent of the company's total data centre revenue, according to the Caixin timeline. In 2022, China had contributed 26.4 per cent of Nvidia's total revenue.
The domestic response
Caixin reports that Huawei captured a 20.4 per cent share of China's AI chip market in 2025 on shipments of 812,000 units, and that internet companies represented nearly two thirds of the country's total AI capital expenditure that year. Semiconductor Manufacturing International Corporation ran at 93.7 per cent capacity utilisation in the second quarter of 2026, with AI-related chip revenue up about 40 per cent from the previous quarter.
Procurement has followed. China Mobile issued a 2.1 billion yuan tender for AI servers in April 2026, and in August 2026 awarded bids for an expansion of its Hohhot intelligent computing centre covering thousands of AI server units built on Huawei's ecosystem, valued at about 1.3 billion yuan. Z.ai began construction of a one gigawatt data centre intended to use domestic chips exclusively, and its GLM-5.3-Flash model relied entirely on homegrown inference chips during testing, according to the report.
Capital markets have moved in step. The five startups Caixin calls the GPU Five Little Tigers, namely Moore Threads, MetaX, Biren, Iluvatar CoreX and Enflame, all pursued public listings during 2026. Newer entrants also raised money. Sunrise launched a cloud inference GPU using LPDDR memory and reached a valuation of 20 billion yuan, while Nano-Core Chip completed a series B round valuing it at 10 billion yuan. DFSX launched an accelerator card built on a domestic 14-nanometre process using a compute-in-memory architecture, and displayed a supernode-based computing cluster at the World AI Conference in Shanghai on 17 July 2026, one day after Huawei's Ascend-based supernode cluster made its offline debut at the same event on 16 July 2026.
What executives and analysts said
Tencent executive Dowson Tong said in June 2026 that no chipmaker had sufficient capacity to meet market demand, and DeepSeek founder Liang Wenfeng identified manufacturing capacity as the core bottleneck for 2026 and 2027, according to Caixin. JPMorgan projected that ByteDance's demand for AI chips would exceed 500,000 units in 2026. These are statements and forecasts from interested parties, not independently verified figures.
What is established and what is merely claimed
Established. Caixin's cover story was published on 31 August 2026. The events in its timeline are documented occurrences: the tightening of US export controls in April 2025, the subscription pauses at Z.ai, Moonshot AI and Alibaba between April and July 2026, SMIC's 93.7 per cent capacity utilisation in the second quarter of 2026, the China Mobile procurement rounds in April and August 2026, and the product showings at the World AI Conference on 16 and 17 July 2026. The 417 per cent demand growth figure originates with the state-backed CAICT and was relayed by Caixin.
Merely claimed or attributed. The causal framing that the blockade sparked the boom is Caixin's editorial interpretation rather than an independently verified finding, since Chinese demand was also rising for domestic reasons. The remarks from Tong and Liang are statements by executives with a commercial stake in the outcome. The JPMorgan figure is a projection, not a measurement. Claims about Z.ai's data centre plans and its model's reliance on domestic chips during testing rest on Caixin's reporting and have not been independently confirmed. Because the full article is subscriber-only, this account is based on the publicly released summary and timeline.
Sources
- Cover Story: How U.S. Tech Blockade Sparked China's AI Chip Boom · Caixin Global